Procedure and House Affairs Committee on Oct. 22nd, 2026

[Christine Normandin, Bloc, Saint-Jean, QC]

Thank you very much.

I'll start with a question for Ms. Crisan.

You talked about the importance of conducting audits on how AI works—for example, on the choices it makes when filtering large volumes of emails. I understand that it requires tremendous human resources to carry out these audits, which are important.

Could you tell us about the best point at which to involve a human in the process? Is it early on—when deciding on a query, for example—that we should ensure the question is properly understood and interpreted by the artificial intelligence? Or should it be at the end of the process, once we have the result, to ensure it meets expectations?

When is the most effective time to carry out a human analysis of the process?

[Dr. Anamaria Crisan]

Thank you for your question.

That's something the research community is also trying to understand. There are certainly ways you could start with a prompt. A bare minimum thing that you could do as a test would be to have a prompt and vary it slightly to see how different the outcome is. This is a known approach that quite a few people use, and that's just before you set anything up. There are some potential mechanisms that you can have downstream, in which you're able to measure the differences: “A prompt like this produces that, and a prompt like that produces this.” There are ways you can do that, and I recommend doing this in order to understand the sensitivity of the summaries or the filtration you're seeing.

What we also see a lot of in research are these “benchmark datasets”. This is as you said: We have a series of scenarios that we decide upon ahead of time, and we test the models against those scenarios to see how they perform. We have a predefined way of judging their accuracy. A lot of benchmarks are built by people who are AI experts or AI engineers and might be auditing their technical capabilities, for example, and how well they perform with a science, math or programming problem.

There are more benchmarks that look at societal questions, such as safety, or the harm of AI systems. People define what is safe, what is harm and what are scenarios when you must act in this or that way. Canada needs a version of this, because there's nobody out there who's saying, “Here's the Canada-specific benchmark that's unique to our demographics and our problems.”

You can do it at both ends. Prompt variation in your office is simple. I would recommend that you give it a try and see what happens. A bigger Canadian values benchmark or audit is more resource-intensive, but it could be valuable to the discussion.

[Christine Normandin, Bloc, Saint-Jean, QC]

Thank you very much.

I have a second question.

You gave the example of hospitals, which still require paper records in the event of a computer system failure. What form of redundancy would be useful in a context where artificial intelligence is increasingly being used? Which systems should be retained by default, as a safeguard?

[Dr. Anamaria Crisan]

It's very simple: critical thinking. It may sound a bit trite, but we're seeing that the more people use these systems.... If, let's say, you use it 100 times, and 99 times or 95% of the time it is getting a good answer, and it fails just once, the question is, how consequential is that failure?

We're seeing a lot of erosion of people's critical thinking and even capacity and capabilities using these systems. I study computer science. We see people not thinking as much about their code, and this can create security vulnerabilities. Do you see what I mean?

The thing is, as people push for more automation and essentially take the human out of the loop, the paper protocol and the safeguard is to insist upon the human in the loop, to insist that we still develop these critical skills as a form of resilience. This is a hard question.

How do we maintain our critical thinking and our autonomy when automation provides useful and reasonable capabilities? It's a challenging question that doesn't have one correct answer.

[Christine Normandin, Bloc, Saint-Jean, QC]

Thank you very much.

Ms. Lister, you mentioned that 74% of electoral bodies use AI, but that only about half of them have a regulatory framework in place to govern its use. You also mentioned that parliaments are best placed to regulate artificial intelligence, for several reasons, including the public's trust in members.

I would like you to tell us about the role of international structures in supporting various national entities that may have gaps or a lack of knowledge. Is there really an interest in having an international framework, given that AI transcends borders?

[Dr. Sarah Lister]

There are different parts of that piece, so I'll perhaps pick off the different parts of your question.

There are international agencies and organizations that are supporting the lesson-learning sharing between parliaments on these issues. The Inter-Parliamentary Union is heavily involved. UNDP is involved. The Commonwealth Parliamentary Association is involved.

I've submitted to the committee and the clerk will share with you a publication I have here that talks about parliaments and artificial intelligence, and it has a list of the types of resources that are available and the types of initiatives. There are also some five or six entities that are tracking different regulatory approaches and so forth.

Lesson learning and consolidation of information, approaches and so on are taking place. The question of international regulation of AI is a very complex one—

[The Chair, Liberal, Chris Bittle]

I'll probably cut you off there. That was excellent.

[Dr. Sarah Lister]

I'm relieved that you have.

[The Chair, Liberal, Chris Bittle]

We'll now turn to Mr. Cooper for five minutes, please.

[Michael Cooper, Conservative, St. Albert—Sturgeon River, AB]

Thank you very much, Mr. Chair.

Thank you to the witnesses.

I'm going to address my first question to Mr. Deveaux.

You noted that, with the advancement of AI, we're receiving a lot of AI inputs, which are likely going to continue to increase in the way of, for example, written submissions to committees and correspondence from constituents.

At the same time, we members and our staff are increasingly using AI to do our jobs, including when our staff uses AI to do research, to look into constituent inquiries and to get suggestions on how to respond to those inquiries. What we're starting to see is an AI loop—AI talking to AI in terms of AI inputs and AI outputs. For me, what is fundamental is that technology augments human judgment and doesn't supplant human judgment.

At what point is there a real risk that, in fact, technology is supplanting human judgment? How do we guard against that as parliamentarians in the work we do?

[Kevin Deveaux]

Well, that's a $64,000 question, isn't it?

As I said, part of it is going to be workload. I go back to my earlier comment on that. I believe that, now, every MP at the federal level gets four staff, or at least the equivalent to four full-time staff. We may need to talk about whether that needs to increase with regard to the research they're doing and the submissions they have to review, whether that might be at the constituency level, committee level or however it looks. I think you're also going to see that a quantity of workers will be required to do this.

Also, it will have to be—going back to another witness's comments—about critical thinking, ensuring that those skills are in place, ensuring that people understand that it's a tool for humans but that humans still have to have the critical thinking skills to discern when...if it's only 1%, as was noted before, of the time.

However, there are issues out there. Just to give you some examples, as I said, in the Australian federal Parliament, going back to 2023 there were submissions to a committee that made accusations against another organization. It turned out that those were hallucinations. They were picked up fairly quickly, mainly because there was an open process. People could read the submission. They identified that a false allegation was put into a particular written submission. We saw from Australia, a few months ago, that the number of hallucinations, the number of false accusations and submissions is increasing. That's going to require, by using a fine-toothed comb, the ability for staff and MPs to identify what is accurate in those submissions.

Also, it's just a human element, in the sense that, as an MP—I say this as a former MLA myself—you need to be in touch with your constituents. It's much as it was in the past. Written submissions would come. There were campaigns to write letters to MPs and committees. You could have a volume of them being submitted. However, there has to be an ability for MPs to be in touch, on the ground, with the people they're engaging with, whether from their critic area portfolio or their constituency, so that they have a nuanced understanding of what is happening and are not necessarily relying just on the volume that might be coming in.

[Michael Cooper, Conservative, St. Albert—Sturgeon River, AB]

You spoke about using AI for legislative drafting, drafting bills, drafting motions. I saw a study that was recently released, from Edinburgh Napier University and Chalmers University of Technology in Sweden, in which research detected thousands of motions in the U.K. and Sweden that were AI-generated or that there were AI inputs in terms of the drafting of those motions. Now, the researchers noted they weren't surprised that this was happening, but they noted that there was not one single disclosure that AI had been used to help draft those legislative products. What are your thoughts on the issue of transparency? Should there be transparency? Is that realistic? We, as legislators, use all kinds of tools. This is just another one, as I see it, but what are your thoughts?

[The Chair, Liberal, Chris Bittle]

Answer very quickly.

[Kevin Deveaux]

It's a good question.

Quickly, I'll say that, around legal drafting, there still needs to be a legal drafter who is signing off on it to form...I believe, in the Canadian system. You're always going to have someone, in the end, who will be endorsing the content of the bill.

With regard to motions, questions and, perhaps, amendments, it's going to be a little looser as a process. I would say that, in those circumstances, there needs to be some recognition and identification that AI was used. There's a difference between saying AI produced something and AI was used as a tool, and the final product is endorsed by the person who's actually submitting it.

[The Chair, Liberal, Chris Bittle]

Thank you so much.

I'll turn to Mr. Jeneroux for five minutes, please.

[Matt Jeneroux, Liberal, Edmonton Riverbend, AB]

Thank you, Mr. Chair.

I want to pick up on that point with you, Mr. Deveaux. The disclosure, in terms of legal drafting, is one thing, but what about the disclosure in terms of speeches that are given in Parliament, communication to constituents and questions in the House? What are your thoughts on whether that particular aspect of disclosure should also be considered?

[Kevin Deveaux]

It's a good question. Let me put it this way: In the last 30 years, how many questions asked in the House were written by a caucus staff member and then handed to an MP? I don't know the statistic on that, but I suspect it's fairly significant. Motions as well might be written by staff. Then it's the MP who's submitting it. What we're doing is swapping out a staff person who may be doing it for AI that is doing it. We don't necessarily declare in a speech when someone gets up and....

As far as I can recall, the rules in most Westminster parliaments are that you're not supposed to read your speeches except for your inaugural speech, and yet that rule is broken practically all the time, I believe. Staff and even individuals are writing their speeches and getting up and reading them. There was an interesting case recently in the Welsh Senedd, the parliament there. Someone was using those new Meta glasses and, they think, reading his speech from his glasses. It looked as though he was talking off the cuff when in fact it was a speech that had been written by him, someone else or AI; we don't know.

This is another issue that will probably be coming up. How do you merge AI with this new sort of gadget technology? How do you ensure that the debates being held in Parliament are authentic and human, based on a discussion and are not just trying to ensure...? There will always be a need for talking points to be put into speeches, but in the end, it should be a debate. It should be a discussion. How do you ensure that it becomes one that is still very much driven by humans and not AI?

[Matt Jeneroux, Liberal, Edmonton Riverbend, AB]

That's interesting. On your comments about increasing staff, I think that's the first we've heard of this as part of the study. Similarly, though, we heard last week that our Hansard is using AI as a first draft for looking through the blues in Parliament. You've identified a lot of the guardrails and risks. Having this for, say, something like Hansard seems to be an opportunity that is presumably going well.

Are there other opportunities you see that could be on the more positive side versus just the riskier side?

[Kevin Deveaux]

Yes. I worry sometimes that we say “AI” as though it's a whole new system. The system before was that someone would speak into a microphone, as we're doing right now, and it would be a voice-to-print system. Even before AI ever came along, there was technology that allowed the editing of Hansard that would originally be a voice-to-print that was done by a computer. I'm happy to see that they're using AI. For me, as someone who searches Hansard on a regular basis, the ability to find that needle in a haystack with regard to particular speeches would be very helpful.

With regard to other areas, I mentioned earlier the vetting of submissions. If we're going to have a proliferation of submissions, then we're probably also going to see AI as a place in which those submissions will be vetted to some extent. There has to be the human element. There need to be some controls. There has to be in place the ability to check and make sure AI is not taking it in the wrong direction with regard to the vetting. I think this is another area we could potentially see.

The third one I mentioned was analysis and the democratization of analysis: the ability of MPs and constituency staff or research staff in a caucus to use AI with bills, particularly these large omnibus bills; the ability for the analysis to be done in a way that can support them; and the ability for them to get information and discern information that will allow them to have a better understanding of the implications of a bill. That's another area in which AI can be useful.

[Matt Jeneroux, Liberal, Edmonton Riverbend, AB]

Ms. Lister, I have less than a minute left. You mentioned how artificial intelligence could improve voter information. I'm wondering if you could quickly share some examples that have been used successfully in other parliaments.

[Dr. Sarah Lister]

I'm happy to share the documents. There'll be some specifics in there in terms of the countries. Many electoral management bodies, for example, are experimenting with chatbots at the time of elections with a closed information source, which helps protect against the misinformation and disinformation we see at the time of elections.

I think we're low on time, but that would be one example.

[The Chair, Liberal, Chris Bittle]

Excellent. We're right at five minutes. That was perfect. Thank you so much.

Ms. Normandin, you have the floor for two and a half minutes.

[Christine Normandin, Bloc, Saint-Jean, QC]

Thank you.

Mr. Deveaux, what do you think of the idea that—before sending a comment, an opinion, a submission or anything else to an MP—people are required to tick a box stating that they have not used AI, that they have used it for research purposes, or that they have used it for drafting purposes, either partially or completely?

It wouldn't prevent them from submitting their document, but it could have some impact. Some people might not want others to know that they haven't thought it through themselves. That wouldn't prevent them from sending their opinion, but it would disclose to some extent the support they may have received from AI.

Do you think that's an avenue worth considering?

[Kevin Deveaux]

That's an interesting question, Madame. My thinking is that it's been three or maybe four years since ChatGPT made it to the public realm, and already the way in which people perceive AI and its use has changed. I think that's going to change continually.

Going back to one of the other witnesses' comments, if we don't maintain critical thinking skills, what we're going to see is that AI will become the dominant source. In 10 years, we will no longer be perhaps talking about whether a person did use AI; it will be a person having to acknowledge they didn't use AI. That will be the surprise.

This goes back to protocols and framework. How do we ensure that we are using the best elements of AI and trying to limit the challenges and the negative aspects of it?

If someone wrote to you now and said, “I used AI for this”, I think there would always be some acknowledgement.... In the end, if you're putting your name on a letter or on a submission, it's yours. It's much like when you, as a member of Parliament, submit a petition: It's your name on the petition you're submitting. Whether it's AI or it's a staff person who helped you draft it, at the moment it won't make much difference, in my mind, with regard to how I would perceive that submission.

In the end, we're probably all going to have to ensure that AI is being used responsibly, so we know that what we're getting is of good quality and is not providing false information.

[The Chair, Liberal, Chris Bittle]

Thank you so much.

Mrs. Block, go ahead for five minutes, please.